A Rare Universal Pattern in Human Languages

Some languages are spoken more quickly than others, but the rate of information they get across is the same.

Two people talk on phones in front of a clock.
Bettmann / Getty

In the early 1960s, a doctoral student at Cornell University wanted to figure out whether there was any truth behind the “cultural stereotype” that certain foreigners speak faster than Americans. He recorded 12 of his fellow students—six Japanese speakers and six American English speakers—monologuing about life on campus, analyzed one minute of each man’s speech, and found that the two groups produced sounds at roughly the same speed. He and a co-author concluded that “the hearer judges the speech rate of a foreign language in terms of his linguistic background,” and that humans the world over were all likely to be more or less equally fast talkers.

In the half century since then, more rigorous studies have shown that, prejudice aside, some languages—such as Japanese, Basque, and Italian—really are spoken more quickly than others. But as mathematical methods and computing power have improved, linguists have spent more time studying not just speech rate, but the effort a speaker has to exert to get a message across to a listener. By calculating how much information every syllable in a language conveys, it’s possible to compare the “efficiency” of different languages. And a study published today in Science Advances found that more efficient languages tend to be spoken more slowly. In other words, no matter how quickly speakers chatter, the rate of information they’re transmitting is roughly the same across languages.

The basic problem of “efficiency,” in linguistics, starts with the trade-off between effort and communication. It takes a certain amount of coordination, and burns a certain number of calories, to make noises come out of your mouth in an intelligible way. And those noises can be more or less informative to a listener, based on how predictable they are. If you and I are discussing dinosaurs, you wouldn’t be surprised to hear me rattle off the names of my favorite species. But if a stranger walks up to you on the street and announces, “Diplodocus!” it’s unexpected. It narrows the scope of possible conversation topics greatly and is therefore highly informative.

Informativity in linguistics is usually calculated per syllable, and it’s measured in bits, just like computer files. The concept can be rather slippery when you’re talking about talking, but essentially, a bit of linguistic information is the amount of information that reduces uncertainty by half. In other words, if I utter a syllable, and that utterance narrows down the set of things I could be talking about from everything in the world to only half the things in the world, that syllable carries one bit of information.

In the new study, the authors calculated the average information density—that is, bits per syllable—of a set of 17 Eurasian languages and compared it with the average speech rate, in syllables per second, of 10 speakers for each language. They found that the rate of information transferred stayed constant—at about 39.15 bits per second, to be exact.

François Pellegrino, the senior author of the new study, says linguists aren’t likely to be surprised to learn that there’s a trade-off between speech rate and information density: “It just confirms what the intuition would be.” But what’s special about his and his team’s work is that, for the first time, they were able “to prove that it holds” for this set of languages.

The speed-efficiency trade-off is likely to be fodder for a long-standing debate among linguists about what language is, and what it’s for. “One of the big divisions in the field of linguistics right now is whether it’s useful to think about language as a code for communication or whether it’s more useful to think about language as something like a mathematical language,” says Richard Futrell, an assistant language-science professor at UC Irvine. This controversy is often described as a fight over whether linguistic universals exist: The language-as-math camp, a.k.a. the generativists, follow in the footsteps of Noam Chomsky and think that certain grammatical rules apply to all languages, while the language-as-communication camp, a.k.a. the functionalists, think that’s bunk.

But the disagreement is more nuanced than that. According to Futrell, functionalists are more likely to get behind research that finds universal truths about how language is used, while Chomskyans are more interested in universals in what language consists of. The relationship between speech rate and information density unearthed in the new study is likely to appeal to functionalists, because it suggests that something about language is fundamentally geared toward transferring information between speakers. But, Futrell says, “I think the Chomskyans would just say this is completely irrelevant.”

The thing about universals is that they have to be true, well, universally. The 17 languages Pellegrino and his colleagues analyzed in this paper are an impressive haul, but they certainly don’t represent the full range of linguistic diversity on Earth, and they completely leave out the languages of Africa, the Americas, and Australia and the Pacific islands. “There’s 6,000 languages in the world,” Futrell says. “So there’s a lot more work to be done in the cross-linguistic comparison.”

The new study also relies on speech-rate data from just 10 speakers per language, all young and educated. Pellegrino isn’t too concerned about his sample size, he says, because the team’s statistical analysis found that the language subjects were speaking had a much larger impact on their speed than their individual identities did. Besides, he says, “cross-linguistic studies using more than a dozen languages are not very common,” and those that do cover such wide ground tend to use only a handful of speakers. “It’s a trade-off in a way as well,” he says.

There’s also the question of whether the speed-efficiency trade-off would hold in spontaneous conversations, not just for text read aloud. Both Pellegrino and Futrell predict that the average information rate for casual speech would be lower than 39 bits per second, but it would still be roughly the same across languages. At least in this select group of 17, exchanging one language for another shouldn’t significantly change the amount of time it takes to get across any given idea. If I had told you about this research in Vietnamese, for example, in the end it probably would have taken just as long.